Anthropic ships Claude Opus 4.7 with stable 1M-token context

Anthropic released Claude Opus 4.7 to all paying tiers this week, promoting the 1M-token context window from last year’s beta to a default capability. Pricing held: $15 per million input, $75 per million output. The announcement matters less for the headline number than for the consistency curve—earlier 1M-context implementations bled accuracy past 200K tokens. The 4.7 release ships internal evals showing flat needle-in-haystack performance across the full window.

For builders, the practical shift is at the architecture layer. RAG pipelines that existed because you couldn’t fit a codebase or contract corpus into context are now a tunable choice rather than a requirement. Several engineering teams are already trimming retrieval steps and replacing them with whole-repo prompts, accepting higher token cost for simpler infrastructure.

Opus 4.7 also lands a quiet improvement to tool use—the model now handles up to 256 parallel tool calls per turn, doubling the previous limit. Combined with the larger context, this nudges the model into a viable foundation for longer-running agent workflows.

Source